Surface fitting using convex tensor-product splines (Q1372094)

The author circumvents the very difficult problem of insuring that a given tensor-product spline surface be convex (or concave) by concentrating on a sufficient condition, insuring that a two-by-two matrix representing the Hessian be definite. He shows that this condition is asymptotically necessary, i.e., in the limit of a continuous surface will describe Dupin's indicatrix. The paper is highly recommended since it gives all details for implementing the procedure (as a quadratic optimization problem). Examples are given to show that even in cases where the surface has to be divided into convex and concave parts, the procedure of approximation by certainly convex or concave parts by the author's method leads to an elimination of all kinds of undesirable wiggles in the tensor surface.